U.S. patent number 10,872,207 [Application Number 16/223,173] was granted by the patent office on 2020-12-22 for determining translation similarity of reverse translations for a plurality of languages.
This patent grant is currently assigned to PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO., LTD.. The grantee listed for this patent is Panasonic Intellectual Property Management Co., Ltd.. Invention is credited to He Cai, Tetsuji Mochida.
![](/patent/grant/10872207/US10872207-20201222-D00000.png)
![](/patent/grant/10872207/US10872207-20201222-D00001.png)
![](/patent/grant/10872207/US10872207-20201222-D00002.png)
![](/patent/grant/10872207/US10872207-20201222-D00003.png)
![](/patent/grant/10872207/US10872207-20201222-D00004.png)
![](/patent/grant/10872207/US10872207-20201222-D00005.png)
![](/patent/grant/10872207/US10872207-20201222-D00006.png)
![](/patent/grant/10872207/US10872207-20201222-D00007.png)
![](/patent/grant/10872207/US10872207-20201222-D00008.png)
![](/patent/grant/10872207/US10872207-20201222-D00009.png)
![](/patent/grant/10872207/US10872207-20201222-D00010.png)
View All Diagrams
United States Patent |
10,872,207 |
Mochida , et al. |
December 22, 2020 |
Determining translation similarity of reverse translations for a
plurality of languages
Abstract
A translation device includes an input unit configured to
receive an input sentence in an original language, a controller
configured to generate a first translation sentence obtained by
translation of the input sentence into a first language, and a
display. The controller generates a second translation sentence
obtained by translation of the input sentence into a second
language different from the first language, a first reverse
translation sentence obtained by reverse translation of the first
translation sentence into the original language, and a second
reverse translation sentence obtained by reverse translation of the
second translation sentence into the original language. The
controller calculates first similarity as similarity between the
first reverse translation sentence and the input sentence, and
determines, in accordance with the first similarity, whether to
display the first translation sentence on the display or to
generate a third translation sentence obtained by translation of
the second reverse translation sentence into the first
language.
Inventors: |
Mochida; Tetsuji (Osaka,
JP), Cai; He (Osaka, JP) |
Applicant: |
Name |
City |
State |
Country |
Type |
Panasonic Intellectual Property Management Co., Ltd. |
Osaka |
N/A |
JP |
|
|
Assignee: |
PANASONIC INTELLECTUAL PROPERTY
MANAGEMENT CO., LTD. (Osaka, JP)
|
Family
ID: |
1000005257727 |
Appl.
No.: |
16/223,173 |
Filed: |
December 18, 2018 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20190213261 A1 |
Jul 11, 2019 |
|
Foreign Application Priority Data
|
|
|
|
|
Jan 9, 2018 [JP] |
|
|
2018-001414 |
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06F
40/58 (20200101); G06F 40/51 (20200101) |
Current International
Class: |
G06F
40/51 (20200101); G06F 40/58 (20200101); G06F
40/47 (20200101); G06F 40/40 (20200101) |
Field of
Search: |
;704/2,3,7,8,277 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
|
|
|
|
|
|
|
2016-157252 |
|
Sep 2016 |
|
JP |
|
2016-218995 |
|
Dec 2016 |
|
JP |
|
6160438 |
|
Jul 2017 |
|
JP |
|
Primary Examiner: Lerner; Martin
Attorney, Agent or Firm: McDermott Will & Emery LLP
Claims
What is claimed is:
1. A translation device comprising: an input unit configured to
receive an input sentence in an original language; a controller
configured to generate a first translation sentence obtained by
translation of the input sentence into a first language; and a
display; wherein the controller: generates a second translation
sentence obtained by translation of the input sentence into a
second language different from the first language, a first reverse
translation sentence obtained by reverse translation of the first
translation sentence into the original language, and a second
reverse translation sentence obtained by reverse translation of the
second translation sentence into the original language, calculates
first similarity as similarity between the first reverse
translation sentence and the input sentence and second similarity
as similarity between the second reverse translation sentence and
the input sentence, displays the first translation sentence on the
display when the first similarity is more than or equal to the
second similarity, generates a third translation sentence obtained
by translation of the second reverse translation sentence into the
first language when the first similarity is less than the second
similarity, generates a fourth translation sentence obtained by
translation of the second reverse translation sentence into the
second language, generates a third reverse translation sentence
obtained by reverse translation of the third translation sentence
into the original language, and a fourth reverse translation
sentence obtained by reverse translation of the fourth translation
sentence into the original language, calculates third similarity as
similarity between the third reverse translation sentence and the
second reverse translation sentence, and fourth similarity as
similarity between the fourth reverse translation sentence and the
second reverse translation sentence, and displays the third
translation sentence on the display when the third similarity is
more than or equal to the fourth similarity.
2. The translation device according to claim 1, wherein the
controller: generates a fifth translation sentence obtained by
translation of the fourth reverse translation sentence into the
first language when the third similarity is less than the fourth
similarity.
3. The translation device according to claim 2, wherein the
controller: generates a sixth translation sentence obtained by
translation of the forth reverse translation sentence into the
second language, generates a fifth reverse translation sentence
obtained by reverse translation of the fifth translation sentence
into the original language, and a sixth reverse translation
sentence obtained by reverse translation of the sixth translation
sentence into the original language, calculates fifth similarity as
similarity between the fifth reverse translation sentence and the
fourth reverse translation sentence, and sixth similarity as
similarity between the sixth reverse translation sentence and the
fourth reverse translation sentence, and displays the fifth
translation sentence on the display when the fifth similarity is
more than or equal to the sixth similarity.
4. The translation device according to claim 1, further comprising
a storage configured to store translation history information
including at least one of the input sentence, the first translation
sentence, the first reverse translation sentence, the second
translation sentence, the second reverse translation sentence, the
first similarity to the fourth similarity, and a number of times of
repeated translation.
5. A translation device comprising: an input unit configured to
receive an input sentence in an original language; a controller
configured to generate a first translation sentence obtained by
translation of the input sentence into a first language; and a
display; wherein the controller: generates a second translation
sentence obtained by translation of the input sentence into a
second language different from the first language, a first reverse
translation sentence obtained by reverse translation of the first
translation sentence into the original language, and a second
reverse translation sentence obtained by reverse translation of the
second translation sentence into the original language, calculates
first similarity as similarity between the first reverse
translation sentence and the input sentence and second similarity
as similarity between the second reverse translation sentence and
the input sentence, displays the first translation sentence on the
display when the first similarity is more than or equal to the
second similarity, generates a third translation sentence obtained
by translation of the second reverse translation sentence into the
first language when the first similarity is less than the second
similarity, generates a fourth translation sentence obtained by
translation of the second reverse translation sentence into the
second language, generates a third reverse translation sentence
obtained by reverse translation of the third translation sentence
into the original language, and a fourth reverse translation
sentence obtained by reverse translation of the fourth translation
sentence into the original language, calculates third similarity as
similarity between the third reverse translation sentence and the
second reverse translation sentence, and fourth similarity as
similarity between the fourth reverse translation sentence and the
second reverse translation sentence, and when the third similarity
is less than the fourth similarity, determines whether or not the
third translation sentence is identical with the first translation
sentence, and displays the first translation sentence on the
display when the third translation sentence is identical with the
first translation sentence.
6. The translation device according to claim 5, wherein the
controller displays the fourth reverse translation sentence along
with the first translation sentence on the display.
7. A translation method comprising: receiving an input sentence in
an original language at an input unit; generating, at a calculator,
a first translation sentence obtained by translation of the input
sentence into a first language, and a second translation sentence
obtained by translation of the input sentence into a second
language different from the first language; generating, at the
calculator, a first reverse translation sentence obtained by
reverse translation of the first translation sentence into the
original language, and a second reverse translation sentence
obtained by reverse translation of the second translation sentence
into the original language; calculating, at the calculator, first
similarity as similarity between the first reverse translation
sentence and the input sentence and second similarity as similarity
between the second reverse translation sentence and the input
sentence; displaying the first translation sentence on a display
when the first similarity is more than or equal to the second
similarity; generating, at the calculator, a third translation
sentence obtained by translation of the second reverse translation
sentence into the first language when the first similarity is less
than the second similarity; generating, at the calculator, a fourth
translation sentence obtained by translation of the second reverse
translation sentence into the second language; generating, at the
calculator, a third reverse translation sentence obtained by
reverse translation of the third translation sentence into the
original language, and a fourth reverse translation sentence
obtained by reverse translation of the fourth translation sentence
into the original language; calculating, at the calculator, third
similarity as similarity between the third reverse translation
sentence and the second reverse translation sentence, and fourth
similarity as similarity between the fourth reverse translation
sentence and the second reverse translation sentence; and
displaying the third translation sentence on the display when the
third similarity is more than or equal to the fourth
similarity.
8. A non-transitory computer-readable recording medium storing a
program causing a computer to execute the translation method
according to claim 7.
Description
BACKGROUND
1. Technical Field
The present disclosure relates to a translation device configured
to translate an input sentence in a language into another language,
a translation method, and a recording medium.
2. Description of the Related Art
Unexamined Japanese Patent Publication No. 2016-157252 discloses a
verification device configured to verify whether or not reverse
translation sentences generated from translation sentences in
different languages are identical with each other. This
verification device generates reverse translation sentences from
translation sentences generated in a plurality of languages by a
plurality of translators. The verification device compares the
plurality of reverse translation sentences and outputs a difference
data list indicating differences. The translators correct the
translation sentences with reference to the difference data list.
This enables efficient check of variations among multilingual
translation sentences to obtain multilingual translation sentences
having no variations.
SUMMARY
The present disclosure provides a translation device and a
translation method achieving accurate translation, as well as a
recording medium.
The translation device according to the present disclosure includes
an input unit configured to receive an input sentence in an
original language, a controller configured to generate a first
translation sentence obtained by translation of the input sentence
into a first language, and a display. The controller generates a
second translation sentence obtained by translation of the input
sentence into a second language different from the first language,
a first reverse translation sentence obtained by reverse
translation of the first translation sentence into the original
language, and a second reverse translation sentence obtained by
reverse translation of the second translation sentence into the
original language. The controller calculates first similarity as
similarity between the first reverse translation sentence and the
input sentence, and determines, in accordance with the first
similarity, whether to display the first translation sentence on
the display or to generate a third translation sentence obtained by
translation of the second reverse translation sentence into the
first language.
These general and specific aspects may be achieved by a system, a
method, a computer program, or any combination among these.
The translation device, the translation method, and the recording
medium according to the present disclosure achieve accurate
translation.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is an outer appearance view of a translation device
according to first and second exemplary embodiments;
FIG. 2 is a block diagram depicting an electric configuration of a
translation system according to the first and second exemplary
embodiments;
FIG. 3 is a view specifically exemplifying translation and reverse
translation according to the first exemplary embodiment;
FIG. 4 is a view exemplifying a display screen of a translation
sentence and a reverse translation sentence according to the first
exemplary embodiment;
FIG. 5 is a flowchart of translation processing according to the
first and second exemplary embodiments;
FIG. 6 is a view exemplifying a selection screen of a target
language (a language spoken by a guest);
FIG. 7 is a flowchart of retranslation processing according to the
first exemplary embodiment;
FIG. 8 is a chart of exemplary translation log information;
FIG. 9 is a view specifically exemplifying translation and reverse
translation according to the second exemplary embodiment;
FIG. 10 is a view exemplifying a display screen having a
translation sentence and a reverse translation sentence according
to the second exemplary embodiment; and
FIG. 11 is a flowchart of retranslation processing according to the
second exemplary embodiment.
DETAILED DESCRIPTION
(Definition of Terms)
The present specification provides definition of terms as
follows.
"original language": language of original sentence
"translation language": language of translation sentence
"target language": language of displayed translation sentence
(language spoken by guest)
"different language": language of translation sentence not
displayed (translation language other than target language)
"intended reverse translation sentence": reverse translation
sentence generated by reverse translation of translation sentence
in target language into original language
"unintended reverse translation sentence": reverse translation
sentence generated by reverse translation of translation sentence
in different language into original language
(Knowledge Configuring Base of the Present Disclosure)
Machine translation from a first language into a second language is
executed in accordance with a translation model obtained through
learning with reference to a bilingual corpus between the first
language and the second language. This bilingual corpus is
generated for each pair of parallel translation languages.
Translation from Japanese into English and reverse translation from
English into Japanese are thus executed in accordance with a
translation model obtained through learning with reference to a
bilingual corpus between Japanese and English. Translation from
Japanese into Chinese and reverse translation from Chinese into
Japanese are executed in accordance with a translation model
obtained through learning with reference to a bilingual corpus
between Japanese and Chinese. Machine translation in accordance
with the translation model achieves lower accuracy as sentences in
the first language and the second language have a larger difference
in construction. Bilingual corpora in some languages may already
include new terms whereas bilingual corpora in other languages may
not yet include such new terms. The bilingual corpus referred to
for learning of the translation model includes translation
sentences that are generated by a plurality of persons and thus
have ununiform translation quality. These have conventionally led
to variation in translation accuracy among languages.
The present disclosure provides a translation device achieving
accurate translation regardless of language types. Specifically,
the present disclosure enables output of a translation sentence,
even in a language exhibiting lower translation accuracy than other
languages, with accuracy as high as accuracy of the other
languages. Specifically, the translation device according to the
present disclosure is configured to generate a translation sentence
obtained by translation of an original sentence into a target
language, and generate a reverse translation sentence (intended
reverse translation sentence) from the translation sentence. The
translation device is further configured to generate a translation
sentence obtained by translation of the original sentence into a
different language, and generate a reverse translation sentence
(unintended reverse translation sentence) from the translation
sentence. When the intended reverse translation sentence has low
accuracy, the translation device generates again a translation
sentence in the target language from the unintended reverse
translation sentence having high accuracy. This configuration
enables output of a correct translation sentence even in a case
where translation into the target language has poor accuracy.
First Exemplary Embodiment
The first exemplary embodiment will be described below with
reference to the drawings.
1. Configuration
1-1. Outer Appearance of Translation Device
FIG. 1 is an outer appearance view of the translation device
according to the present exemplary embodiment. FIG. 1 depicts
translation device 1 of a tablet type configured to translate
conversation between two users who speak different languages.
Translation device 1 is configured to execute translation from a
first language into a second language (translation language) as
well as translation from the second language into the first
language. The present exemplary embodiment will be described below
assuming that translation device 1 translates face-to-face
conversation between a host (e.g. guide) speaking the first
language and a guest (e.g. traveler) speaking the second language.
The present exemplary embodiment adopts Japanese as the first
language spoken by the host.
Translation device 1 includes microphone 10, speaker unit 12,
display 14, and touch panel 16. Microphone 10 and speaker unit 12
are exemplarily disposed near openings in a side surface of
translation device 1. Display 14 and touch panel 16 are disposed on
a primary surface of translation device 1. Display 14 has an area
adjacent to the host and near a first end in a longitudinal
direction, and the area includes utterance icon 14h and display
area 15h. Display 14 has an area adjacent to the guest and near a
second end in the longitudinal direction, and the area includes
utterance icon 14g and display area 15g. Utterance icons 14h, 14g
accept touch operation by the users.
Utterance icon 14h is an operation icon operated by the host to
specify start and end points of utterance when the host gives the
utterance, in other words, when the host inputs utterance in the
first language to translation device 1. Utterance icon 14g is an
operation icon operated by the guest to specify start and end
points of utterance when the guest gives the utterance, in other
words, when the guest inputs utterance in the second language.
Display areas 15h, 15g are provided for display of an original
sentence, a translation sentence, a reverse translation sentence,
or the like in the form of a character string.
1-2. Configurations of Translation Device and Server
FIG. 2 is a block diagram depicting an electric configuration of a
translation system according to the present exemplary embodiment.
The present disclosure provides translation system 100 including
translation device 1 depicted in FIG. 1, voice recognition server
3, translation server 4, and voice synthesis server 5. Translation
device 1 executes data communication with voice recognition server
3, translation server 4, and voice synthesis server 5, via network
2 like the Internet.
Voice recognition server 3 receives digital voice data from
translation device 1 via network 2 and executes voice recognition
of the digital voice data thus received to generate text data of an
original sentence (voice recognition data in the form of a
character string).
Translation server 4 is equipped with a plurality of translation
engines 40. Translation server 4 according to the present exemplary
embodiment includes translation engines 40a, 40b, 40c, 40d.
Translation engine 40a executes translation from Japanese into
English and reverse translation from English into Japanese in
accordance with a translation model obtained through learning with
reference to a bilingual corpus between Japanese and English.
Translation engine 40b executes translation from Japanese into
Chinese and reverse translation from Chinese into Japanese in
accordance with a translation model obtained through learning with
reference to a bilingual corpus between Japanese and Chinese.
Translation engine 40c executes translation from Japanese into
Korean and reverse translation from Korean into Japanese in
accordance with a translation model obtained through learning with
reference to a bilingual corpus between Japanese and Korean.
Translation engine 40d executes translation from Japanese into Thai
and reverse translation from Thai into Japanese in accordance with
a translation model obtained through learning with reference to a
bilingual corpus between Japanese and Thai. Translation server 4 is
equipped with the plurality of translation engines 40a, 40b, 40c,
40d for achievement of translation from Japanese into the plurality
of languages. Translation engine 40a, 40b, 40c, 40d according to
the present application will be referred to as translation engine
40 in a case where translation engines 40a, 40b, 40c, 40d are not
particularly distinguished from one another. When translation
server 4 receives the text data of the original sentence from
translation device 1 via network 2, translation engine 40 generates
text data of a translation sentence (translation data in the form
of a character string) from the text data of the original sentence
thus received. When translation server 4 receives the text data of
the translation sentence from translation device 1 via network 2,
translation engine 40 generates text data of a reverse translation
sentence (reverse translation data in the form of a character
string) obtained by reverse translation of the text data of the
translation sentence thus received.
Voice synthesis server 5 receives the text data of the translation
sentence from translation device 1 via network 2 and executes voice
synthesis of the text data of the translation sentence thus
received to generate a voice signal.
Translation device 1 includes microphone 10, speaker unit 12,
display 14, touch panel 16, as well as communication unit 18,
storage 20, and controller 22.
Microphone 10 is a device configured to convert voice to digital
voice data. Specifically, microphone 10 converts voice in a voice
signal (analog electric signal), and further converts the voice
signal to digital voice data with use of an AD converter.
Microphone 10 exemplarily embodies an input unit used for input of
an original sentence to translation device 1. The input unit is not
limited to microphone 10, but may include a keyboard, a button, a
switch, a camera, or any combination thereof. Examples of the
original sentence include a sentence according to utterance (voice)
and a sentence directly input via the input unit.
Communication unit 18 is a communication module configured to
execute data communication with voice recognition server 3,
translation server 4, and voice synthesis server 5 via network 2 in
accordance with a communication system such as Bluetooth
(registered trademark), Wi-Fi (registered trademark), 3G, long term
evolution (LTE) (registered trademark), or the institute of
electrical and electronics engineers (IEEE) 802.11.
Storage 20 is a recording medium including a flash memory, a
ferroelectric memory, a hard disk drive (HDD), a solid state drive
(SSD), a random access memory (RAM), or any combination thereof.
Storage 20 stores digital voice data, text data of an original
sentence, text data of a translation sentence, and text data of a
reverse translation sentence. Storage 20 also stores various
programs for controller 22.
Controller 22 includes a central processing unit (CPU), a micro
processing unit (MPU), or the like and executes the various
programs stored in storage 20 to control entire behavior of
translation device 1. Controller 22 reads out data or a program
stored in storage 20 and executes various calculation processing
for achievement of a predetermined function. The function of
controller 22 according to the present exemplary embodiment is
achieved through cooperation between hardware and software.
Controller 22 may alternatively be embodied only with a hardware
circuit dedicatedly designed for achievement of a predetermined
function. Specifically, controller 22 may include, the CPU or the
MPU, alternatively a digital signal processor (DSP), a
field-programmable gate array (FPGA), an application specific
integrated circuit (ASIC), or the like.
Speaker unit 12 is configured to convert an electric signal to
voice. Speaker unit 12 outputs voice according to the voice signal
(electric signal) transmitted from controller 22. Speaker unit 12
exemplarily embodies an output unit used for output of voice or
text indicating a translation result.
Display 14 is configured to display an image, and is configured by
a liquid crystal display device or an organic electroluminescence
(EL) display device. Display 14 displays, in display area 15h, 15g,
an image indicated by text data of an original sentence, text data
of a translation sentence, or text data of a reverse translation
sentence transmitted from controller 22. Display 14 exemplarily
embodies a display configured to present, to the host or the guest,
the text data of the original sentence, the text data of the
translation sentence, or the text data of the reverse translation
sentence. Display 14 also displays utterance icons 14h, 14g
mentioned above.
Touch panel 16 exemplarily embodies an operation unit to be
operated by the user and accepts a command from the user. Touch
panel 16 is superimposed on display 14. The operation unit in
translation device 1 is not limited to touch panel 16, but may
include a keyboard, a button, a switch, a camera, or any
combination thereof. Touch panel 16 may replace microphone 10 to
function as the input unit used for input of an original sentence
to translation device 1.
2. Behavior
2-1. Outlined Translation and Reverse Translation
Outlined translation and reverse translation will be described with
reference to FIG. 3 and FIG. 4. FIG. 3 exemplifies translation and
reverse translation in a case where the target language (a language
spoken by the guest) is Thai. FIG. 4 exemplifies display of a
translation sentence and a reverse translation sentence obtained
through retranslation processing. The present exemplary embodiment
includes repeated translation of an unintended reverse translation
sentence until an intended reverse translation sentence becomes
similar to an input sentence or a reinput sentence.
As depicted in FIG. 3, when the host speaks "Morning" in Japanese,
translation system 100 translates input sentence 51 (original
sentence) "Morning" into Thai as the target language, as well as
into different languages including English, Chinese, and Korean.
Translation system 100 executes reverse translation of translation
sentence 52d in Thai into Japanese as the original language to
generate intended reverse translation sentence 53d. Translation
system 100 executes reverse translation of translation sentences
52a, 52b, 52c in English, Chinese, and Korean to generate
unintended reverse translation sentences 53a, 53b, 53c. FIG. 3
exemplifies intended reverse translation sentence 53d "Yes, I can
take it (in Japanese)", and unintended reverse translation sentence
53a, 53b "Good morning (in Japanese)" generated from the
translation sentences in English and Chinese. Intended reverse
translation sentence 53d "Yes, I can take it (in Japanese)" is not
similar to input sentence 51 "Morning (in Japanese)". In contrast,
unintended reverse translation sentences 53a, 53b "Good morning (in
Japanese)" are similar to input sentence 51 "Morning (in
Japanese)". This indicates that translation sentence 52d in Thai is
highly possibly incorrect. Translation system 100 thus executes
translation again.
Translation system 100 newly generates translation sentences 55a,
55b, 55c in English, Chinese, and Korean, as well as translation
sentence 55d in Thai, from unintended reverse translation sentence
53a or unintended reverse translation sentence 53b "Good morning
(in Japanese)", which is similar to the input sentence and serves
as reinput sentence 54. Translation system 100 then newly generates
intended reverse translation sentence 56d from translation sentence
55d in Thai, and newly generates unintended reverse translation
sentences 56a, 56b, 56c from translation sentences 55a, 55b, 55c in
English, Chinese, and Korean. FIG. 3 exemplifies newly generated
intended reverse translation sentence 56d "Good morning (in
Japanese)". Intended reverse translation 56d "Good Morning (in
Japanese)" is similar to reinput sentence 54 "Good Morning (in
Japanese)" substantially equally to unintended reverse translation
sentences 56a, 56b "Good morning (in Japanese)" generated from
translation sentences 55a, 55b in English and Chinese. Intended
reverse translation 56d "Good Morning (in Japanese)" is more
similar to reinput sentence 54 "Good Morning (in Japanese)" than
unintended reverse translation sentences 56c "Hello (in Japanese)"
generated from translation sentence 55c in Korean. This indicates
that translation sentence 55d in Thai is correct. Translation
device 1 thus displays, on display 14, translation sentence 55d in
Thai obtained by translation of reinput sentence 54 "Good morning
(in Japanese)" and intended reverse translation sentence 56d
corresponding to translation sentence 55d. As exemplarily depicted
in FIG. 4, translation device 1 displays translation sentence 55d
in Thai as the target language in display area 15g adjacent to the
guest and displays intended reverse translation sentence 56d in
display area 15h adjacent to the host.
2-2. Translation Processing
Behavior of translation system 100 will be described with reference
to FIG. 5 to FIG. 7. FIG. 5 depicts behavior of translation
processing executed by controller 22 in translation device 1. FIG.
6 is a view exemplifying selection screen 60 for the target
language (a language spoken by the guest). FIG. 7 depicts
retranslation processing (details of step S9 in FIG. 5) executed by
controller 22 in translation device 1.
With reference to FIG. 5, controller 22 in translation device 1
initially sets a target language in accordance with selection by
the user (S1). As depicted in FIG. 6, translation device 1
displays, on display 14, language selection icons 61 for selection
of the language spoken by the guest before the host and guest start
conversation, for example. Languages selectable with use of
language selection icons 61 are dependent on the languages to be
translated by translation engines 40a, 40b, 40c, 40d. Prior to
conversation, the guest touches an appropriate one of language
selection icons 61 to select the language spoken by the guest.
Controller 22 sets the target language in accordance with language
selection icon 61 having been touched. In an exemplary case where
the guest touches language selection icon 61 for English,
controller 22 sets English as the target language.
Controller 22 acquires an input sentence (original sentence) (S2).
Specifically, controller 22 receives digital voice data indicating
utterance (voice) of the host via microphone 10. Controller 22
transmits the digital voice data to voice recognition server 3 via
network 2. Voice recognition server 3 executes voice recognition of
the digital voice data thus received to generate text data of the
original sentence as a spoken sentence. Controller 22 receives the
text data of the original sentence from voice recognition server 3
via network 2.
Controller 22 generates translation sentences in the target
language and the different languages from the original sentence
serving as the input sentence (S3). Specifically, controller 22
transmits the text data of the input sentence (original sentence)
to translation server 4 via network 2. Translation server 4 causes
translation engines 40a, 40b, 40c, 40d to generate text data of
translation sentences from the text data of the original sentence
thus received. Translation server 4 transmits the text data of the
translation sentences to translation device 1 via network 2.
Translation device 1 receives the text data of the translation
sentences from translation server 4 via network 2.
Controller 22 translates the translation sentences into the
original language to generate reverse translation sentences (S4).
In this case, controller 22 generates an intended reverse
translation sentence from the translation sentence in the target
language, and unintended reverse translation sentences from the
translation sentences in the different languages. Specifically,
controller 22 transmits the text data of the translation sentences
to translation server 4 via network 2. Translation server 4 causes
translation engines 40a, 40b, 40c, 40d to generate text data of
reverse translation sentences from the text data of the translation
sentences. Translation server 4 transmits the text data of the
reverse translation sentences to translation device 1 via network
2. Translation device 1 receives the text data of the reverse
translation sentences from translation server 4 via network 2.
Controller 22 calculates similarity between the input sentence and
each of the reverse translation sentences (S5). Controller 22
calculates the similarity in accordance with a technique such as
Word2vec, neural network, or Levenshtein distance. Controller 22
compares the similarity between the intended reverse translation
sentence and the input sentence with the similarity between each of
the unintended reverse translation sentences and the input sentence
(S6).
Controller 22 determines whether the similarity between the
intended reverse translation sentence and the input sentence is
more than or equal to the similarity between each of the unintended
reverse translation sentences and the input sentence (S7). In this
case, controller 22 compares the similarity of the intended reverse
translation sentence with the similarity of each of the unintended
reverse translation sentences. FIG. 3 exemplifies the case where
controller 22 compares similarity of intended reverse translation
sentence 53d generated from translation sentence 52d in Thai with
similarity of each of unintended reverse translation sentences 53a,
53b, 53c generated from translation sentences 52a, 52b, 52c in
English, Chinese, and Korean. If the similarity of the intended
reverse translation sentence is more than or equal to the
similarity of each of the unintended reverse translation sentences
(Yes in S7), controller 22 outputs the translation sentence in the
target language and the intended reverse translation sentence (S8).
For example, controller 22 displays the translation sentence in the
target language in display area 15g adjacent to the guest and
displays the intended reverse translation sentence in display area
15h adjacent to the host. Controller 22 may output voice indicating
a translation sentence upon displaying the translation sentence. In
this case, controller 22 transmits the text data of the translation
sentences to voice synthesis server 5 via network 2. Voice
synthesis server 5 executes voice synthesis in accordance with the
text data of each of the translation sentences received from
translation device 1 to generate a voice signal. Voice synthesis
server 5 transmits the voice signal to translation device 1 via
network 2. Controller 22 outputs, from speaker unit 12, voice
indicating a translation result in accordance with the voice signal
received from voice synthesis server 5.
If the similarity of the intended reverse translation sentence is
less than the similarity of each of the unintended reverse
translation sentences (No in S7), controller 22 executes
retranslation processing (S9). Specifically, if at least one of the
unintended reverse translation sentences has similarity higher than
the similarity of the intended reverse translation sentence,
controller 22 executes the retranslation processing.
2-3. Retranslation Processing
The retranslation processing (details of step S9 in FIG. 5) will be
described with reference to FIG. 7. Controller 22 newly generates
translation sentences in the target language and the different
languages (S901). Controller 22 compares the similarity of the
intended reverse translation sentence to the input sentence with
the similarity of each of the unintended reverse translation
sentences to the input sentence. Controller 22 regards, as a
reinput sentence, the unintended reverse translation sentence
having similarity higher than the similarity of the intended
reverse translation sentence. For example, controller 22 regards,
as the reinput sentence, the unintended reverse translation
sentence having the highest similarity. Generation of a translation
sentence from a reinput sentence is executed in accordance with a
method identical with the method of generating a translation
sentence from an input sentence (step S3 in FIG. 5). Specifically,
controller 22 transmits text data of the reinput sentence to
translation server 4 to acquire text data of translation sentences
from translation server 4.
Controller 22 newly generates an intended reverse translation
sentence from the translation sentence in the target language, and
newly generates unintended reverse translation sentences from the
translation sentences in the different languages (S902). Generation
of a reverse translation sentence in the retranslation processing
is executed in accordance with a method identical with the method
of generating a reverse translation sentence in the translation
processing (step S4 in FIG. 5). Specifically, controller 22
transmits the text data of the translation sentences to translation
server 4 to acquire text data of the reverse translation sentences
from translation server 4.
Controller 22 calculates similarity between the reinput sentence
and each of the reverse translation sentences (S903). Controller 22
calculates the similarity in accordance with a technique such as
Word2vec, neural network, or Levenshtein distance, as in step S5 in
FIG. 5. Controller 22 compares the similarity between the intended
reverse translation sentence and the reinput sentence with the
similarity between each of the unintended reverse translation
sentences and the reinput sentence (S904). Controller 22 determines
whether the similarity between the intended reverse translation
sentence and the reinput sentence is more than or equal to the
similarity between each of the unintended reverse translation
sentences and the reinput sentence (S905).
If the similarity of the intended reverse translation sentence is
more than or equal to the similarity of each of the unintended
reverse translation sentences (Yes in S905), controller 22 displays
the translation sentence in the target language and the intended
reverse translation sentence (S906). As exemplarily depicted in
FIG. 4, controller 22 displays translation sentence 55d in the
target language in display area 15g adjacent to the guest and
displays intended reverse translation sentence 56d in display area
15h adjacent to the host. If the similarity of the intended reverse
translation sentence is less than the similarity of each of the
unintended reverse translation sentences (No in S905), in other
words, if at least one of the unintended reverse translation
sentences has similarity more than the similarity of the intended
reverse translation sentence, controller 22 determines whether or
not the number of times of translation has exceeded a predetermined
number of times (S907). If the number of times of translation has
not yet exceeded the predetermined number of times (No in S907),
controller 22 returns to step S901 to repeat generation of
translation sentences and reverse translation sentences from, as a
new reinput sentence, the unintended reverse translation sentence
having similarity to the reinput sentence higher than the
similarity of the intended reverse translation sentence.
If the number of times of translation has exceeded the
predetermined number of times (Yes in S907), controller 22
displays, on display 14, a sentence indicating impossibility of
translation (S908). For example, controller 22 displays the
sentence indicating impossibility of translation in display area
15h adjacent to the host.
3. Effects, Supplementals, Etc.
Translation device 1 according to the present exemplary embodiment
includes microphone 10 (exemplifying the input unit) configured to
receive input of an original sentence in an original language,
controller 22 configured to generate a translation sentence (first
translation sentence) obtained by translation of the original
sentence as an input sentence into a target language (first
language), and display 14 (exemplifying the display) configured to
display the translation sentence. Controller 22 generates a
translation sentence (second translation sentence) obtained by
translation of the input sentence into at least one language
(second language) different from the target language, an intended
reverse translation sentence (first reverse translation sentence)
obtained by reverse translation of the translation sentence in the
target language into the original language, and an unintended
reverse translation sentence (second reverse translation sentence)
obtained by reverse translation of the translation sentence in the
different language into the original language. Controller 22
calculates similarity (first similarity) between the intended
reverse translation sentence and the input sentence (S5).
Controller 22 determines, in accordance with the similarity,
whether to display the translation sentence in the target language
on display 14 or to newly generate a translation sentence (third
translation sentence) obtained by translation of the unintended
reverse translation sentence as a reinput sentence into the target
language (S6, S7).
Specifically, controller 22 calculates similarity (second
similarity) between the unintended reverse translation sentence and
the input sentence, and displays, on display 14, the translation
sentence in the target language if the similarity between the
intended reverse translation sentence and the input sentence is
more than or equal to the similarity between the unintended reverse
translation sentence and the input sentence. If the similarity
between the intended reverse translation sentence and the input
sentence is less than the similarity between the unintended reverse
translation sentence and the input sentence, controller 22 newly
generates a translation sentence (third translation sentence)
obtained by translation of the unintended reverse translation
sentence as a reinput sentence into the target language.
If translation into the target language has low accuracy, a
translation sentence is to be newly generated from the unintended
reverse translation sentence that is generated from the translation
sentence in the different language having high translation
accuracy. This enables output of a highly accurate translation
sentence. Accurate translation is thus achieved regardless of the
type of language. The unintended reverse translation sentence is
generated by translation engine 40 and is thus more appropriate for
translation than the original sentence. The unintended reverse
translation sentence is to be translated more accurately than the
original sentence. Retranslation of the unintended reverse
translation sentence reduces the number of times of repeated
utterance or repeated input by the user.
The configuration further achieves the following effects. The
present exemplary embodiment achieves higher translation accuracy
with a larger number of different translation languages. The
translation sentence in the target language is generated from the
unintended reverse translation sentence, so that accurate
translation is achieved with reference to the bilingual corpus in
the different language even in a case where the bilingual corpus in
the target language does not include a new term. Accurate
translation is enabled without addition of the new term to every
bilingual corpus, which leads to reduction in cost for preparation
of the bilingual corpora. A language exhibiting low translation
accuracy can be specified in accordance with the number of times of
repeated translation. This leads to more efficient improvement of
the translation model. An incorrectly translated sentence can be
specified for every language. This also leads to more efficient
improvement of the translation model.
Furthermore, controller 22 newly generates a translation sentence
(fourth translation sentence) obtained by translation of the
reinput sentence into at least one different language. Controller
22 newly generates an intended reverse translation sentence (third
reverse translation sentence) obtained by reverse translation of
the newly generated translation sentence in the target language
into the original language, and an unintended reverse translation
sentence (exemplifying a fourth reverse translation sentence)
obtained by reverse translation of the newly generated translation
sentence in the different language into the original language.
Controller 22 calculates similarity (third similarity) between the
newly generated intended reverse translation sentence and the
reinput sentence as well as similarity (fourth similarity) between
the newly generated unintended reverse translation sentence and the
reinput sentence (S903). If the similarity between the intended
reverse translation sentence and the reinput sentence is more than
or equal to the similarity between the unintended reverse
translation sentence and the reinput sentence, controller 22
displays, on display 14, the newly generated translation sentence
in the target language. If the similarity between the intended
reverse translation sentence and the reinput sentence is less than
the similarity between the unintended reverse translation sentence
and the reinput sentence, controller 22 repeats generation of a
translation sentence in the target language, an intended reverse
translation sentence, a translation sentence in the different
language, and an unintended reverse translation sentence, from the
newly generated unintended reverse translation sentence as the
reinput sentence. Specifically, controller 22 generates a
translation sentence (fifth translation sentence) obtained by
translation of the unintended reverse translation sentence into the
target language.
If translation into the target language has poor accuracy, a
translation sentence in the target language is repeatedly generated
from the unintended reverse translation sentence that is generated
from the translation sentence in the different language having high
translation accuracy. This enables output of a highly accurate
translation sentence.
Controller 22 may generate translation log information 24
(exemplifying translation history information) indicating
translation records during execution of the translation processing
in FIG. 5 and the retranslation processing in FIG. 7, and store
translation log information 24 in storage 20. FIG. 8 depicts
exemplary translation log information 24. Translation log
information 24 exemplarily includes all or at least one of the
original sentence (input sentence), the number of times of
translation, the reinput sentence, the translation sentence in the
target language, the intended reverse translation sentence,
similarity between the intended reverse translation sentence and
the input sentence (or the reinput sentence), the translation
sentence in the different language, the unintended reverse
translation sentence, and similarity between the unintended reverse
translation sentence and the input sentence (or the reinput
sentence). Controller 22 may determine the reinput sentence with
reference to translation log information 24.
Second Exemplary Embodiment
The present exemplary embodiment enables output of a correct
translation sentence even in a case where a language includes a
polyseme. In such a language including a polyseme, a single term
has a plurality of meanings and an input sentence may not agree
with a reverse translation sentence even if translation is executed
correctly. For example, Korean has a single greeting term for "Good
morning", "Hello", and "Good evening". In this case, as depicted in
FIG. 9, even with correct translation sentences 72c, 75c in Korean
for input sentence 71 "Morning (in Japanese)" and reinput sentence
74 "Good morning (in Japanese)", reverse translation sentences 73c,
76c may have "Hello (in Japanese)". Repeated translation may thus
fail to improve similarity between the reverse translation sentence
and the reinput sentence, in which case no translation sentence may
be output. The present exemplary embodiment achieves output of any
correct translation sentence even in a case where a reverse
translation sentence is different from an input sentence or a
reinput sentence in a language including a single term having a
plurality of meanings. Translation system 100 according to the
present exemplary embodiment is configured identically with the
translation system in FIG. 2 according to the first exemplary
embodiment.
Outlined translation and reverse translation according to the
second exemplary embodiment will be described with reference to
FIG. 9 and FIG. 10. FIG. 9 exemplifies translation and reverse
translation in a case where the target language (a language spoken
by the guest) is Korean. FIG. 10 depicts an exemplary display
screen of a translation sentence and a reverse translation sentence
according to the second exemplary embodiment.
As depicted in FIG. 9, when the host speaks "Morning (in
Japanese)", translation system 100 translates input sentence 71
(original sentence) "Morning (in Japanese)" into Korean as the
target language, as well as into different languages including
English, Chinese, and Thai. Translation system 100 further
generates intended reverse translation sentence 73c from
translation sentence 72c in Korean, and generates unintended
reverse translation sentences 73a, 73b, 73d from translation
sentences 72a, 72b, 72d in English, Chinese, and Thai. Intended
reverse translation sentence 73c "Hello (in Japanese)" and input
sentence 71 "Morning (in Japanese)" have similarity less than
similarity between each of unintended reverse translation sentences
73a, 73b "Good morning (in Japanese)" and input sentence 71
"Morning". Translation system 100 thus executes translation
again.
Translation system 100 newly generates translation sentences 75a,
75b, 75c, 75d in English, Chinese, Korean, and Thai, from
unintended reverse translation sentence 73a or unintended reverse
translation sentence 73b "Good morning (in Japanese)", which is
similar to input sentence 71 and serves as reinput sentence 74.
Translation system 100 also generates intended reverse translation
sentence 76c from translation sentence 75c in Korean, and newly
generates unintended reverse translation sentences 76a, 76b, 76d
from translation sentences 75a, 75b, 75d in English, Chinese, and
Thai. Intended reverse translation sentence 76c "Hello (in
Japanese)" and reinput sentence 74 "Good morning (in Japanese)"
have similarity less than similarity between each of unintended
reverse translation sentences 76a, 76b, 76d "Good morning (in
Japanese)" and reinput sentence 74 "Good morning". However,
previous translation sentence 72c and current translation sentence
75c in the target language agree with each other. In this case,
translation system 100 determines that translation sentences 72c,
75c in the target language are correct, assuming that a polyseme
causes disagreement between intended reverse translation sentence
76c and reinput sentence 74. Translation system 100 thus displays,
on display 14, translation sentence 72c (translation sentence 75c)
in Korean. Translation system 100 also displays, on display 14,
unintended reverse translation sentence 76a, unintended reverse
translation sentence 76b, or unintended reverse translation
sentence 76d "Good morning (in Japanese)" having high similarity to
reinput sentence 74, in place of intended reverse translation
sentence 76c. For example, display area 15g for the guest has the
translation sentence in Korean as the target language, and display
area 15h for the host has the reverse translation sentence from
English, Chinese, or Thai "Good morning (in Japanese)" having the
highest similarity. As exemplarily depicted in FIG. 10, display
area 15g may have translation sentence 72c in Korean and display
area 15h may have unintended reverse translation sentence 76a from
English. Even in the case where intended reverse translation
sentence 76c "Hello (in Japanese)" generated from translation
sentence 75c in Korean and reinput sentence 74 "Good morning (in
Japanese)" have low similarity, if previously generated translation
sentence 72c in Korean is identical with currently generated
translation sentence 75c in Korean, translation sentences 72c, 75c
are determined as being correct. Translation sentence 72c
(translation sentence 75c) is presented on display 14.
Translation device 1 will be described in terms of behavior in
retranslation processing according to the second exemplary
embodiment with reference to FIG. 11. FIG. 11 depicts the
retranslation processing (details of step S9 in FIG. 5) according
to the second exemplary embodiment.
Step S901 to step S908 in FIG. 11 are identical with step S901 to
step S908 in FIG. 7 according to the first exemplary embodiment.
FIG. 11 depicts step S951 to step S953 in addition to the steps
depicted in FIG. 7. Described below are further added step S951 to
step S953.
The process flow according to the present exemplary embodiment
proceeds to step S951 if similarity between the intended reverse
translation sentence generated from the translation sentence in the
target language and the reinput sentence is lower than similarity
between the unintended reverse translation sentence generated from
the translation sentence in the different language and the reinput
sentence (No in S905). Controller 22 compares the previous
translation sentence and the current translation sentence in the
target language (S951), and determines whether or not the previous
translation sentence and the current translation sentence in the
target language are identical with each other (S952). If the
previous translation sentence and the current translation sentence
in the target language are identical with each other (Yes in S952),
controller 22 displays the translation sentence in the target
language as well as the unintended reverse translation sentence
having the highest similarity to the reinput sentence (S953). If
the previous translation sentence and the current translation
sentence in the target language are different from each other (No
in S952), controller 22 determines whether or not the number of
times of translation has exceeded a predetermined number of times
(S907).
As described above, if the newly generated intended reverse
translation sentence (first reverse translation sentence) and the
reinput sentence have similarity (third similarity) lower than
similarity (fourth similarity) between the newly generated
unintended reverse translation sentence (second reverse translation
sentence) and the reinput sentence, controller 22 determines
whether or not the newly generated translation sentence in the
target language is identical with the previously generated
translation sentence in the target language (S952). If the newly
generated translation sentence in the target language is identical
with the previously generated translation sentence in the target
language, controller 22 displays, on display 14, the translation
sentence in the target language (S953). This achieves accurate
output of a translation sentence even when a polyseme is
involved.
Controller 22 displays, on display 14, the translation sentence in
the target language as well as the unintended reverse translation
sentence having similarity to the reinput sentence higher than the
similarity of the intended reverse translation sentence. This
allows the host to check whether or not the original sentence is
translated correctly.
Other Exemplary Embodiments
The exemplary embodiments have been described above to exemplify
the techniques disclosed in the present application. The techniques
according to the present disclosure are not limited by the
exemplary embodiments, and are applicable to other exemplary
embodiments achieved through change, replacement, addition,
elimination, or the like where appropriate. Such other exemplary
embodiments will be described below.
The above exemplary embodiments include step S7 in FIG. 5 and step
S905 in FIG. 7 and FIG. 11 of determining whether or not the
similarity of the intended reverse translation sentence is more
than or equal to the similarity of the unintended reverse
translation sentence for display of the translation sentence in the
target language or execution of the retranslation processing. Step
S7 or step S905 may include different determination. Alternatively,
whether or not the similarity of the intended reverse translation
sentence is more than a threshold (e.g. a minimum score value) may
be determined for execution of the retranslation processing when
the similarity is less than or equal to the threshold. Still
alternatively, assuming "a first threshold (e.g. the minimum score
value)<a second threshold", the retranslation processing may be
executed if the similarity of the intended reverse translation
sentence is less than or equal to the first threshold and the
similarity of the unintended reverse translation sentence is more
than or equal to the second threshold. Further alternatively, the
retranslation processing may be executed if the similarity of the
intended reverse translation sentence is less than or equal to a
threshold (e.g. the minimum score value) and the similarity of the
intended reverse translation sentence and the similarity of the
unintended reverse translation sentence have a difference more than
or equal to a predetermined value.
According to the above exemplary embodiments, step S903 in FIG. 7
and FIG. 11 relates to calculation of similarity between the
reinput sentence and each reverse translation sentence. Similarity
between the input sentence (original sentence) and each reverse
translation sentence may alternatively be calculated in addition or
in place of this calculation.
The above exemplary embodiments include repeating translation in
step S901 in FIG. 7 and FIG. 11 from the unintended reverse
translation sentence having similarity to the reinput sentence
higher (e.g. highest among the unintended reverse translation
sentences) than the similarity of the intended reverse translation
sentence. The reinput sentence may be determined according to a
different criterion. Alternatively, the reinput sentence may be
regarded as the unintended reverse translation sentence generated
most recently and having the highest similarity to the original
sentence. Still alternatively, the reinput sentence may be selected
from the unintended reverse translation sentences generated most
recently as well as in the past, to have similarity to the original
sentence higher than the similarity of the intended reverse
translation sentence.
The above exemplary embodiments include step S907 in FIG. 7 and
FIG. 11 of determining whether or not to repeat the retranslation
processing in accordance with whether or not the number of times of
translation has exceeded the predetermined number of times. This
determination may alternatively be made in accordance with a
different criterion. For example, determination may be made as to
whether or not there is any reverse translation sentence not yet
used as a reinput sentence among the unintended reverse translation
sentences having similarity to the reinput sentence or the original
sentence more than or equal to a threshold, and the retranslation
processing may be executed repeatedly if there is any such reverse
translation sentence. The selected unintended reverse translation
sentence having the similarity to the reinput sentence or the
original sentence more than or equal to the threshold is not
limited to the currently translated sentences but may be selected
from past unintended reverse translation sentences generated in all
the translation processing. Determination may be made as to whether
or not the similarity of the unintended reverse translation
sentence to the reinput sentence or the original sentence is less
than similarity of the immediately previous unintended reverse
translation sentence, and the retranslation processing may be
stopped if the similarity is less than the immediately previous
similarity.
If the previous translation sentence and the current translation
sentence in the target language are identical with each other in
the determination in step S952 in FIG. 11 and the unintended
reverse translation sentence and the reinput sentence (or the
original sentence) have similarity more than the threshold, the
process flow may proceed to step S953 for display of the
translation sentence in the target language. If the previous
translation sentence and the current translation sentence in the
target language are identical with each other and the similarity of
the unintended reverse translation sentence to the reinput sentence
and the similarity to the original sentence have a difference
within a predetermined value, the process flow may proceed to step
S953 for display of the translation sentence in the target
language.
Translation device 1 according to the above exemplary embodiments
is configured to translate an original sentence according to
utterance (voice) input via microphone 10. Translation device 1 may
alternatively be configured to translate an original sentence not
in the form of utterance but input via the keyboard or a mouse, or
an original sentence generated in accordance with gesture in a sign
language or the like captured with use of a camera.
Translation device 1 according to the above exemplary embodiments
is configured to calculate similarity between a reverse translation
sentence and an input sentence or a reinput sentence. Such
calculation of similarity may alternatively be executed by
translation server 4. For example, translation server 4 may
generate a translation sentence and a reverse translation sentence
from the input sentence or the reinput sentence, calculate
similarity, and repeat translation until the similarity of the
intended reverse translation sentence exceeds the similarity of the
unintended reverse translation sentence. In this case, translation
server 4 may transmit the translation sentence in the target
language and the intended reverse translation sentence to
translation device 1 when the similarity of the intended reverse
translation sentence becomes more than or equal to the similarity
of the unintended reverse translation sentence. If the number of
times of translation exceeds the predetermined number of times
before the similarity of the intended reverse translation sentence
becomes more than or equal to the similarity of the unintended
reverse translation sentence, translation server 4 may transmit
information indicating impossibility of translation to translation
device 1. Translation device 1 displays, on display 14, the
translation sentence in the target language and the intended
reverse translation sentence, which are acquired from translation
server 4.
In the above exemplary embodiments, voice recognition server 3
executes voice recognition, translation server 4 executes
translation, and voice synthesis server 5 executes voice synthesis.
However, the present disclosure is not limited to the exemplary
embodiments. Translation device 1 can alternatively execute at
least one of voice recognition, translation, and voice synthesis.
For example, translation device 1 (terminal) may have all the
functions of voice recognition server 3, translation server 4, and
voice synthesis server 5 to be configured to execute all processing
relevant to translation. Translation device 1 may not include
communication unit 18 in this case. Voice recognition, translation,
and voice synthesis may be executed by one or two servers.
The above exemplary embodiments exemplify the case where there are
three different languages. It is required to be at least one
different language. Translation server 4 including the plurality of
translation engines is configured to execute translation and
reverse translation among the plurality of languages. The
translation server including a single translation engine may
alternatively be configured to execute translation and reverse
translation among the plurality of languages.
The above exemplary embodiments exemplify the case where the
original language is Japanese and the translation languages include
English, Chinese, Korean, and Thai. The original language and the
translation languages are not limited to those according to the
above exemplary embodiments. Examples of the original language and
the translation languages may alternatively include any other
language (e.g. German, French, Spanish, Vietnamese, or
Indonesian).
(Outline of Exemplary Embodiments)
(1) The translation device according to the present disclosure
includes an input unit configured to receive an input sentence in
an original language, a controller configured to generate a first
translation sentence obtained by translation of the input sentence
into a first language, and a display. The controller generates a
second translation sentence obtained by translation of the input
sentence into a second language different from the first language,
a first reverse translation sentence (intended reverse translation
sentence) obtained by reverse translation of the first translation
sentence into the original language, and a second reverse
translation sentence (unintended reverse translation sentence)
obtained by reverse translation of the second translation sentence
into the original language, and calculates first similarity as
similarity between the first reverse translation sentence and the
input sentence (S5). The controller determines, in accordance with
the first similarity, whether to display the first translation
sentence on the display or to generate a third translation sentence
obtained by translation of the second reverse translation sentence
into the first language (S6, S7).
Translation is executed again from the second reverse translation
sentence in accordance with the similarity to the input sentence,
to achieve accurate translation regardless of language types.
(2) The controller in the translation device according to (1) may
calculate second similarity as similarity between the second
reverse translation sentence and the input sentence. The controller
may display the first translation sentence on the display if the
first similarity is more than or equal to the second similarity,
and may generate the third translation sentence if the first
similarity is less than the second similarity.
(3) The controller in the translation device according to (2) may
generate a fourth translation sentence obtained by translation of
the second reverse translation sentence into the second language, a
third reverse translation sentence (intended reverse translation
sentence) obtained by reverse translation of the third translation
sentence into the original language, and a fourth reverse
translation sentence (unintended reverse translation sentence)
obtained by reverse translation of the fourth translation sentence
into the original language. The controller may calculate third
similarity as similarity between the third reverse translation
sentence and the second reverse translation sentence, and fourth
similarity as similarity between the fourth reverse translation
sentence and the second reverse translation sentence (S903). The
controller may display the third translation sentence on the
display if the third similarity is more than or equal to the fourth
similarity, and may generate a fifth translation sentence obtained
by translation of the fourth reverse translation sentence into the
first language if the third similarity is less than the fourth
similarity.
Translation is repeated until similarity to the reinput sentence
improves, for achievement of accurate translation regardless of
language types.
(4) The controller in the translation device according to (2) may
generate the fourth translation sentence obtained by translation of
the second reverse translation sentence into the second language,
the third reverse translation sentence obtained by reverse
translation of the third translation sentence into the original
language, and the fourth reverse translation sentence obtained by
reverse translation of the fourth translation sentence into the
original language. The controller may calculate the third
similarity as similarity between the third reverse translation
sentence and the second reverse translation sentence, and the
fourth similarity as similarity between the fourth reverse
translation sentence and the second reverse translation sentence.
The controller may determine whether or not the third translation
sentence is identical with the first translation sentence (S952) if
the third similarity is less than the fourth similarity, and may
display the first translation sentence on the display if the third
translation sentence is identical with the first translation
sentence.
This configuration achieves display of an accurate translation
sentence even in a case where the target language has a
polyseme.
(5) The controller in the translation device according to (4) may
display, on the display, the first translation sentence as well as
the fourth reverse translation sentence.
This configuration achieves display of a reverse translation
sentence appropriate for the input sentence or the reinput sentence
even in a case where the target language has a polyseme. This
allows the user to find whether or not the translation sentence is
correct by checking the reverse translation sentence.
(6) The translation device according to (3) may further include a
storage configured to store translation history information
including at least one of the input sentence, the first translation
sentence, the first reverse translation sentence, the second
translation sentence, the second reverse translation sentence, the
first similarity to the fourth similarity, and the number of times
of repeated translation.
(7) The present disclosure provides a translation method including:
receiving an input sentence in an original language at an input
unit (S2); generating, at a calculator (controller 22 and
translation server 4 included in translation device 1), a first
translation sentence obtained by translation of the input sentence
into a first language, and a second translation sentence obtained
by translation of the input sentence into a second language
different from the first language (S3); generating, at the
calculator (controller 22 and translation server 4 in translation
device 1), a first reverse translation sentence obtained by reverse
translation of the first translation sentence into the original
language, and a second reverse translation sentence obtained by
reverse translation of the second translation sentence into the
original language (S4); calculating, at the calculator (controller
22), similarity between the first reverse translation sentence and
the input sentence (S5); and determining, at the calculator
(controller 22), in accordance with the similarity, whether to
display the first translation sentence on a display or to generate
a third translation sentence obtained by translation of the second
reverse translation sentence into the first language (S6, S7).
The translation device or the translation method recited in
appropriate one of the claims of the present disclosure is achieved
by cooperation between a program and a hardware resource such as a
processor or a memory.
The present disclosure is applicable to a translation device
configured to execute translation according to voice of a speaker
or an input sentence.
* * * * *